def generate_rho0(self):
"""
Generate list of density normalizations for the number
of trials given at initialization. Based on Schoedel
et al. (2009), assuming M_BH = 4.1x10^6 Msun, gamma=1.0
"""
rho0ave = 3.2e4 # solar masses/pc^3
rho0sig = 1.3e4 # solar masses/pc^3
rho0 = zeros(self.ntrials, dtype=float64)
gens = aorb.create_generators(1, self.ntrials)
rho0gen = gens[0]
for ii in range(self.ntrials):
rho0[ii] = rho0gen.gauss(rho0ave, rho0sig)
self.rho0 = rho0
def generate_rho0(self):
"""
Generate list of density normalizations for the number
of trials given at initialization. Based on Schoedel
et al. (2009), assuming M_BH = 4.1x10^6 Msun, gamma=1.0
"""
rho0ave = 3.2e4 # solar masses/pc^3
rho0sig = 1.3e4 # solar masses/pc^3
rho0 = zeros(self.ntrials, dtype=float64)
gens = aorb.create_generators(1, self.ntrials)
rho0gen = gens[0]
for ii in range(self.ntrials):
rho0[ii] = rho0gen.gauss(rho0ave, rho0sig)
self.rho0 = rho0
def generate(self, efitFile=None):
"""
Generate the list of masses, distances, and focus positions
for the number of trials specified at initialization. The
resulting variables are arrays of mass, distance, x0, and y0:
m -- mass in solar masses
r0 -- distance in pc
x0 -- X origin in pixels
y0 -- Y origin in pixels
There are two ways to generate this list:
1) Use the output from Andrea's efit mass/Ro monte carlos. The
number of trials spit out by efit must be the same or larger
than the number of trials specified here. To use this option,
set the parameter 'efitFile' to the name of the file containing
the efit monte carlo results.
2) Use a gaussian distribution of mass/Ro/focus position. To
specify this option (this is the default) just use efitFile=None
The distributions are determined by the variables:
self.massCenter # solar masses
self.massSigma # solar masses
self.distCenter # parsec
self.distSigma # parsec
self.x0Center # pixel
self.x0Sigma # pixel
self.y0Center # pixel
self.y0Sigma # pixel
which should be set after initialization but before calling
generate(). If no values are specified, the default values from
the Eisenhauer (2006) paper will be used.
"""
m = zeros(self.ntrials, dtype=float64)
r0 = zeros(self.ntrials, dtype=float64)
x0 = zeros(self.ntrials, dtype=float64)
y0 = zeros(self.ntrials, dtype=float64)
if (efitFile == None):
gens = aorb.create_generators(4, self.ntrials)
mgen = gens[0]
r0gen = gens[1]
x0gen = gens[2]
y0gen = gens[3]
for ii in range(self.ntrials):
m[ii] = mgen.gauss(self.massCenter, self.massSigma)
r0[ii] = r0gen.gauss(self.distCenter, self.distSigma)
x0[ii] = x0gen.gauss(self.x0Center, self.x0Sigma)
y0[ii] = y0gen.gauss(self.y0Center, self.y0Sigma)
else:
# Read in M/Ro simluation results
# Columns are:
# Dist x0 y0 a P e t0 w i Ome
# pc pix pix mas yr yr
lines = open(efitFile, 'r').readlines()
if (self.ntrials > len(lines)):
print('EFIT file too short for number of trials = ',
len(lines))
for ii in range(self.ntrials):
fields = lines[ii].split()
r0[ii] = float(fields[0]) # in pc
x0[ii] = float(fields[1]) # in pixels
y0[ii] = float(fields[2]) # in pixels
# Get the semi-major axis and period in order to get the mass
a = float(fields[3]) # in mas
p = float(fields[4]) # in years
a *= r0[ii] / 1000.0
# Mass in Msun
m[ii] = a**3 / p**2
self.m = m
self.r0 = r0
self.x0 = x0
self.y0 = y0
def generate(self, efitFile=None):
"""
Generate the list of masses, distances, and focus positions
for the number of trials specified at initialization. The
resulting variables are arrays of mass, distance, x0, and y0:
m -- mass in solar masses
r0 -- distance in pc
x0 -- X origin in pixels
y0 -- Y origin in pixels
There are two ways to generate this list:
1) Use the output from Andrea's efit mass/Ro monte carlos. The
number of trials spit out by efit must be the same or larger
than the number of trials specified here. To use this option,
set the parameter 'efitFile' to the name of the file containing
the efit monte carlo results.
2) Use a gaussian distribution of mass/Ro/focus position. To
specify this option (this is the default) just use efitFile=None
The distributions are determined by the variables:
self.massCenter # solar masses
self.massSigma # solar masses
self.distCenter # parsec
self.distSigma # parsec
self.x0Center # pixel
self.x0Sigma # pixel
self.y0Center # pixel
self.y0Sigma # pixel
which should be set after initialization but before calling
generate(). If no values are specified, the default values from
the Eisenhauer (2006) paper will be used.
"""
m = zeros(self.ntrials, dtype=float64)
r0 = zeros(self.ntrials, dtype=float64)
x0 = zeros(self.ntrials, dtype=float64)
y0 = zeros(self.ntrials, dtype=float64)
if efitFile == None:
gens = aorb.create_generators(4, self.ntrials)
mgen = gens[0]
r0gen = gens[1]
x0gen = gens[2]
y0gen = gens[3]
for ii in range(self.ntrials):
m[ii] = mgen.gauss(self.massCenter, self.massSigma)
r0[ii] = r0gen.gauss(self.distCenter, self.distSigma)
x0[ii] = x0gen.gauss(self.x0Center, self.x0Sigma)
y0[ii] = y0gen.gauss(self.y0Center, self.y0Sigma)
else:
# Read in M/Ro simluation results
# Columns are:
# Dist x0 y0 a P e t0 w i Ome
# pc pix pix mas yr yr
lines = open(efitFile, "r").readlines()
if self.ntrials > len(lines):
print "EFIT file too short for number of trials = ", len(lines)
for ii in range(self.ntrials):
fields = lines[ii].split()
r0[ii] = float(fields[0]) # in pc
x0[ii] = float(fields[1]) # in pixels
y0[ii] = float(fields[2]) # in pixels
# Get the semi-major axis and period in order to get the mass
a = float(fields[3]) # in mas
p = float(fields[4]) # in years
a *= r0[ii] / 1000.0
# Mass in Msun
m[ii] = a ** 3 / p ** 2
self.m = m
self.r0 = r0
self.x0 = x0
self.y0 = y0
def run():
# Assume circular orbits with isotropic normal vectors on circular
# orbits with circular orbital velocity of 500 km/s.
ntrials = 10000
nstars = int((1.0 / 3.0) * 100)
gens = aorb.create_generators(7, ntrials*10)
velTot = 500.0 # km/s -- velocity amplitude
radTot = 2.04 # arcsec
def fitfunc(p, fjac=None, vx=None, vy=None, vz=None,
vxerr=None, vyerr=None, vzerr=None):
irad = math.radians(p[0])
orad = math.radians(p[1])
nx = math.sin(irad) * math.cos(orad)
ny = -math.sin(irad) * math.sin(orad)
nz = -math.cos(irad)
top = (nx*vx + ny*vy + nz*vz)**2
bot = (nx*vxerr + ny*vyerr + nz*vzerr)**2
devs = (1.0 / (len(vx) - 1.0)) * top / bot
status = 0
return [status, devs.flat]
# Keep track from every trial the incl, omeg, chi2, number of stars
incl = na.zeros(ntrials, type=na.Float)
omeg = na.zeros(ntrials, type=na.Float)
chi2 = na.zeros(ntrials, type=na.Float)
niter = na.zeros(ntrials)
stars = na.zeros(ntrials)
# Keep track of same stuff for spatial test
inclPos = na.zeros(ntrials, type=na.Float)
omegPos = na.zeros(ntrials, type=na.Float)
chi2Pos = na.zeros(ntrials, type=na.Float)
niterPos = na.zeros(ntrials)
angleAvg = na.zeros(ntrials, type=na.Float)
angleStd = na.zeros(ntrials, type=na.Float)
for trial in range(ntrials):
if ((trial % 100) == 0):
print 'Trial %d' % trial
x = na.zeros(nstars, type=na.Float)
y = na.zeros(nstars, type=na.Float)
z = na.zeros(nstars, type=na.Float)
vx = na.zeros(nstars, type=na.Float)
vy = na.zeros(nstars, type=na.Float)
vz = na.zeros(nstars, type=na.Float)
rmag = na.zeros(nstars, type=na.Float)
vmag = na.zeros(nstars, type=na.Float)
nx = na.zeros(nstars, type=na.Float)
ny = na.zeros(nstars, type=na.Float)
nz = na.zeros(nstars, type=na.Float)
for ss in range(nstars):
vx[ss] = gens[0].uniform(-velTot, velTot)
vyTot = math.sqrt(velTot**2 - vx[ss]**2)
vy[ss] = gens[1].uniform(-vyTot, vyTot)
vz[ss] = math.sqrt(velTot**2 - vx[ss]**2 - vy[ss]**2)
vz[ss] *= gens[2].choice([-1.0, 1.0])
x[ss] = gens[3].uniform(-radTot, radTot)
yTot = math.sqrt(radTot**2 - x[ss]**2)
y[ss] = gens[4].uniform(-yTot, yTot)
z[ss] = math.sqrt(radTot**2 - x[ss]**2 - y[ss]**2)
z[ss] *= gens[5].choice([-1.0, 1.0])
rmag[ss] = math.sqrt(x[ss]**2 + y[ss]**2 + z[ss]**2)
vmag[ss] = math.sqrt(vx[ss]**2 + vy[ss]**2 + vz[ss]**2)
rvec = [x[ss], y[ss], z[ss]]
vvec = [vx[ss], vy[ss], vz[ss]]
tmp = util.cross_product(rvec, vvec)
tmp /= rmag[ss] * vmag[ss]
nx[ss] = tmp[0]
ny[ss] = tmp[1]
nz[ss] = tmp[2]
r2d = na.hypot(x, y)
v2d = na.hypot(vx, vy)
top = (x * vy - y * vx)
jz = (x * vy - y * vx) / (r2d * v2d)
djzdx = (vy * r2d * v2d - (top * v2d * x / r2d)) / (r2d * v2d)**2
djzdy = (-vx * r2d * v2d - (top * v2d * y / r2d)) / (r2d * v2d)**2
djzdvx = (-y * r2d * v2d - (top * r2d * vx / v2d)) / (r2d * v2d)**2
djzdvy = (x * r2d * v2d - (top * r2d * vy / v2d)) / (r2d * v2d)**2
xerr = na.zeros(nstars, type=na.Float) + 0.001 # arcsec
yerr = na.zeros(nstars, type=na.Float) + 0.001
vxerr = na.zeros(nstars, type=na.Float) + 10.0 # km/s
vyerr = na.zeros(nstars, type=na.Float) + 10.0
vzerr = na.zeros(nstars, type=na.Float) + 30.0 # km/s
jzerr = na.sqrt((djzdx*xerr)**2 + (djzdy*yerr)**2 +
(djzdvx*vxerr)**2 + (djzdvy*vyerr)**2)
# Eliminate all stars with jz > 0 and jz/jzerr < 2
# I think these are they cuts they are doing
idx = (na.where((jz < 0) & (na.abs(jz/jzerr) > 2)))[0]
#idx = (na.where(jz < 0))[0]
#idx = range(len(jz))
cotTheta = vz / na.sqrt(vx**2 + vy**2)
phi = na.arctan2(vy, vx)
# Initial guess:
p0 = na.zeros(2, type=na.Float)
p0[0] = gens[5].uniform(0.1, 90) # deg -- inclination
p0[1] = gens[6].uniform(0.1, 360) # deg -- omega
# Setup properties of each free parameter.
parinfo = {'relStep':10.0, 'step':10.0, 'fixed':0,
'limits':[0.0,360.0],
'limited':[1,1], 'mpside':1}
pinfo = [parinfo.copy() for i in range(len(p0))]
pinfo[0]['limits'] = [0.0, 180.0]
pinfo[1]['limits'] = [0.0, 360.0]
# Stuff to pass into the fit function
functargs = {'vx': vx[idx], 'vy': vy[idx], 'vz': vz[idx],
'vxerr':vxerr[idx], 'vyerr':vyerr[idx], 'vzerr':vzerr[idx]}
m = mpfit.mpfit(fitfunc, p0, functkw=functargs, parinfo=pinfo,
quiet=1)
if (m.status <= 0):
print 'error message = ', m.errmsg
p = m.params
incl[trial] = p[0]
omeg[trial] = p[1]
stars[trial] = len(idx)
chi2[trial] = m.fnorm / (stars[trial] - len(p0))
niter[trial] = m.niter
n = [math.sin(p[0]) * math.cos(p[1]),
-math.sin(p[0]) * math.sin(p[1]),
-math.cos(p[0])]
# Now look at the angle between the best fit normal vector
# from the velocity data and the true r cross v normal vector.
# Take the dot product between n and nreal.
angle = na.arccos(n[0]*nx + n[1]*ny + n[2]*nz)
angle *= (180.0 / math.pi)
# What is the average angle and std angle
angleAvg[trial] = angle.mean()
angleStd[trial] = angle.stddev()
# print chi2[trial], chi2Pos[trial], incl[trial], inclPos[trial], \
# omeg[trial], omegPos[trial], niter[trial], niterPos[trial]
# print angleAvg[trial], angleStd[trial]
# Plot up chi2 for v-fit vs. chi2 for x-fit
pylab.clf()
pylab.semilogx(chi2, angleAvg, 'k.')
pylab.errorbar(chi2, angleAvg, fmt='k.', yerr=angleStd)
pylab.xlabel('Chi^2')
pylab.ylabel('Angle w.r.t. Best Fit')
foo = raw_input('Contine?')
# Probability of encountering solution with chi^2 < 2
idx = (na.where(chi2 < 2.0))[0]
print 'Prob(chi^2 < 2) = %5.3f ' % (len(idx) / float(ntrials))
# Probability of encountering solution with chi^2 < 2 AND
# inclination = 20 - 30 and Omega = 160 - 170
foo = (na.where((chi2 < 2.0) & (incl > 20) & (incl < 40)))[0]
print 'Prob of chi2 and incl = %5.3f' % (len(foo) / float(ntrials))
pylab.clf()
pylab.subplot(2, 2, 1)
pylab.hist(chi2, bins=na.arange(0, 10, 0.5))
pylab.xlabel('Log Chi^2')
pylab.subplot(2, 2, 2)
pylab.hist(incl[idx])
pylab.xlabel('Inclination for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 180, rng[2], rng[3]])
pylab.subplot(2, 2, 3)
pylab.hist(omeg[idx])
pylab.xlabel('Omega for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 360, rng[2], rng[3]])
pylab.subplot(2, 2, 4)
pylab.hist(stars[idx])
pylab.xlabel('Nstars for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 33, rng[2], rng[3]])
pylab.savefig('diskTest.png')
# Pickle everything
foo = {'incl': incl, 'omeg': omeg, 'star': stars,
'chi2': chi2, 'niter': niter}
pickle.dump(foo, open('diskTestSave.pick', 'w'))
def run():
# Assume circular orbits with isotropic normal vectors on circular
# orbits with circular orbital velocity of 500 km/s.
ntrials = 10000
nstars = int((1.0 / 3.0) * 100)
gens = aorb.create_generators(7, ntrials*10)
velTot = 500.0 # km/s -- velocity amplitude
radTot = 2.04 # arcsec
def fitfunc(p, fjac=None, vx=None, vy=None, vz=None,
vxerr=None, vyerr=None, vzerr=None):
irad = math.radians(p[0])
orad = math.radians(p[1])
nx = math.sin(irad) * math.cos(orad)
ny = -math.sin(irad) * math.sin(orad)
nz = -math.cos(irad)
top = (nx*vx + ny*vy + nz*vz)**2
bot = (nx*vxerr + ny*vyerr + nz*vzerr)**2
devs = (1.0 / (len(vx) - 1.0)) * top / bot
status = 0
return [status, devs.flat]
# Keep track from every trial the incl, omeg, chi2, number of stars
incl = na.zeros(ntrials, type=na.Float)
omeg = na.zeros(ntrials, type=na.Float)
chi2 = na.zeros(ntrials, type=na.Float)
niter = na.zeros(ntrials)
stars = na.zeros(ntrials)
# Keep track of same stuff for spatial test
inclPos = na.zeros(ntrials, type=na.Float)
omegPos = na.zeros(ntrials, type=na.Float)
chi2Pos = na.zeros(ntrials, type=na.Float)
niterPos = na.zeros(ntrials)
angleAvg = na.zeros(ntrials, type=na.Float)
angleStd = na.zeros(ntrials, type=na.Float)
for trial in range(ntrials):
if ((trial % 100) == 0):
print('Trial %d' % trial)
x = na.zeros(nstars, type=na.Float)
y = na.zeros(nstars, type=na.Float)
z = na.zeros(nstars, type=na.Float)
vx = na.zeros(nstars, type=na.Float)
vy = na.zeros(nstars, type=na.Float)
vz = na.zeros(nstars, type=na.Float)
rmag = na.zeros(nstars, type=na.Float)
vmag = na.zeros(nstars, type=na.Float)
nx = na.zeros(nstars, type=na.Float)
ny = na.zeros(nstars, type=na.Float)
nz = na.zeros(nstars, type=na.Float)
for ss in range(nstars):
vx[ss] = gens[0].uniform(-velTot, velTot)
vyTot = math.sqrt(velTot**2 - vx[ss]**2)
vy[ss] = gens[1].uniform(-vyTot, vyTot)
vz[ss] = math.sqrt(velTot**2 - vx[ss]**2 - vy[ss]**2)
vz[ss] *= gens[2].choice([-1.0, 1.0])
x[ss] = gens[3].uniform(-radTot, radTot)
yTot = math.sqrt(radTot**2 - x[ss]**2)
y[ss] = gens[4].uniform(-yTot, yTot)
z[ss] = math.sqrt(radTot**2 - x[ss]**2 - y[ss]**2)
z[ss] *= gens[5].choice([-1.0, 1.0])
rmag[ss] = math.sqrt(x[ss]**2 + y[ss]**2 + z[ss]**2)
vmag[ss] = math.sqrt(vx[ss]**2 + vy[ss]**2 + vz[ss]**2)
rvec = [x[ss], y[ss], z[ss]]
vvec = [vx[ss], vy[ss], vz[ss]]
tmp = util.cross_product(rvec, vvec)
tmp /= rmag[ss] * vmag[ss]
nx[ss] = tmp[0]
ny[ss] = tmp[1]
nz[ss] = tmp[2]
r2d = na.hypot(x, y)
v2d = na.hypot(vx, vy)
top = (x * vy - y * vx)
jz = (x * vy - y * vx) / (r2d * v2d)
djzdx = (vy * r2d * v2d - (top * v2d * x / r2d)) / (r2d * v2d)**2
djzdy = (-vx * r2d * v2d - (top * v2d * y / r2d)) / (r2d * v2d)**2
djzdvx = (-y * r2d * v2d - (top * r2d * vx / v2d)) / (r2d * v2d)**2
djzdvy = (x * r2d * v2d - (top * r2d * vy / v2d)) / (r2d * v2d)**2
xerr = na.zeros(nstars, type=na.Float) + 0.001 # arcsec
yerr = na.zeros(nstars, type=na.Float) + 0.001
vxerr = na.zeros(nstars, type=na.Float) + 10.0 # km/s
vyerr = na.zeros(nstars, type=na.Float) + 10.0
vzerr = na.zeros(nstars, type=na.Float) + 30.0 # km/s
jzerr = na.sqrt((djzdx*xerr)**2 + (djzdy*yerr)**2 +
(djzdvx*vxerr)**2 + (djzdvy*vyerr)**2)
# Eliminate all stars with jz > 0 and jz/jzerr < 2
# I think these are they cuts they are doing
idx = (na.where((jz < 0) & (na.abs(jz/jzerr) > 2)))[0]
#idx = (na.where(jz < 0))[0]
#idx = range(len(jz))
cotTheta = vz / na.sqrt(vx**2 + vy**2)
phi = na.arctan2(vy, vx)
# Initial guess:
p0 = na.zeros(2, type=na.Float)
p0[0] = gens[5].uniform(0.1, 90) # deg -- inclination
p0[1] = gens[6].uniform(0.1, 360) # deg -- omega
# Setup properties of each free parameter.
parinfo = {'relStep':10.0, 'step':10.0, 'fixed':0,
'limits':[0.0,360.0],
'limited':[1,1], 'mpside':1}
pinfo = [parinfo.copy() for i in range(len(p0))]
pinfo[0]['limits'] = [0.0, 180.0]
pinfo[1]['limits'] = [0.0, 360.0]
# Stuff to pass into the fit function
functargs = {'vx': vx[idx], 'vy': vy[idx], 'vz': vz[idx],
'vxerr':vxerr[idx], 'vyerr':vyerr[idx], 'vzerr':vzerr[idx]}
m = mpfit.mpfit(fitfunc, p0, functkw=functargs, parinfo=pinfo,
quiet=1)
if (m.status <= 0):
print('error message = ', m.errmsg)
p = m.params
incl[trial] = p[0]
omeg[trial] = p[1]
stars[trial] = len(idx)
chi2[trial] = m.fnorm / (stars[trial] - len(p0))
niter[trial] = m.niter
n = [math.sin(p[0]) * math.cos(p[1]),
-math.sin(p[0]) * math.sin(p[1]),
-math.cos(p[0])]
# Now look at the angle between the best fit normal vector
# from the velocity data and the true r cross v normal vector.
# Take the dot product between n and nreal.
angle = na.arccos(n[0]*nx + n[1]*ny + n[2]*nz)
angle *= (180.0 / math.pi)
# What is the average angle and std angle
angleAvg[trial] = angle.mean()
angleStd[trial] = angle.stddev()
# print chi2[trial], chi2Pos[trial], incl[trial], inclPos[trial], \
# omeg[trial], omegPos[trial], niter[trial], niterPos[trial]
# print angleAvg[trial], angleStd[trial]
# Plot up chi2 for v-fit vs. chi2 for x-fit
pylab.clf()
pylab.semilogx(chi2, angleAvg, 'k.')
pylab.errorbar(chi2, angleAvg, fmt='k.', yerr=angleStd)
pylab.xlabel('Chi^2')
pylab.ylabel('Angle w.r.t. Best Fit')
foo = input('Contine?')
# Probability of encountering solution with chi^2 < 2
idx = (na.where(chi2 < 2.0))[0]
print('Prob(chi^2 < 2) = %5.3f ' % (len(idx) / float(ntrials)))
# Probability of encountering solution with chi^2 < 2 AND
# inclination = 20 - 30 and Omega = 160 - 170
foo = (na.where((chi2 < 2.0) & (incl > 20) & (incl < 40)))[0]
print('Prob of chi2 and incl = %5.3f' % (len(foo) / float(ntrials)))
pylab.clf()
pylab.subplot(2, 2, 1)
pylab.hist(chi2, bins=na.arange(0, 10, 0.5))
pylab.xlabel('Log Chi^2')
pylab.subplot(2, 2, 2)
pylab.hist(incl[idx])
pylab.xlabel('Inclination for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 180, rng[2], rng[3]])
pylab.subplot(2, 2, 3)
pylab.hist(omeg[idx])
pylab.xlabel('Omega for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 360, rng[2], rng[3]])
pylab.subplot(2, 2, 4)
pylab.hist(stars[idx])
pylab.xlabel('Nstars for Chi^2 < 2')
rng = pylab.axis()
pylab.axis([0, 33, rng[2], rng[3]])
pylab.savefig('diskTest.png')
# Pickle everything
foo = {'incl': incl, 'omeg': omeg, 'star': stars,
'chi2': chi2, 'niter': niter}
pickle.dump(foo, open('diskTestSave.pick', 'w'))